The General Election isn't really and election - it's 650 elections. First Past the Post will explore what happens when you aggregate 650 constituency betting markets into one forecast.
Our model is simple. We turn bookmakers' odds into probabilities and use these to simulate the election. There are five components:
Odds. We get these from the bookmakers. Odds describe the payoff of a winning bet.
Implied probability. The way odds are expressed as probability: odds to the power of -1.
Re-weighted probability. Implied probabilities sum to more than 100% - this is how bookmakers make a profit. We re-weight probabilities so that they sum to 100%.
Control for long-shot bias. Bettors as a group have a tendency to overvalue long-shots. We compensate by removing candidates with a probability of less than 20%.
Monte Carlo simulation. Generates a random percentage value between 0 and 100%. When the random value is less than or equal to a candidate's win-probability, that candidate wins. We calculate this 10,000 times for each candidate in each seat; the average outcome is our forecast.